Another thing that's missing is the lifetime expectancy, e.g. "for how many years does something encrypted in 2030 need to be unbreakable?"
The author doesn't seem to be a big authority, so has little to lose by staking their reputation on "you don't need it to be that good," whereas by the very nature of their authority, anyone in the resource you link is going to be motivated to never be wrong under any circumstances. So if someone with some reputation/authority/power to lose think there's a 0.001% chance that some new incremental improvements will allow for fast-enough breaking of 2048 bit encryption created in 2030 within a window where that would be unacceptable, then they're motivated to guess high. The authority in this case doesn't directly bear the costs of too high of a guess, whereas it could be very bad for, i dunno, some country's government, and by extension the org or people that made that country's standards recommendations, if some classified information became public 15 or 50 years earlier than intended just because it could be decrypted.